基于AM-CNN-LSTM模型的柴油机NO_x排放预测

刘星¹，周圣凯¹，田淋瑕¹，邓小超¹，林鹏慧¹，刘泽都²，雷艳^2*

1.广西玉柴机器股份有限公司，广西 玉林  537000；

2.北京工业大学机械与能源工程学部汽车系，北京  100124

摘要：为精确控制选择性催化还原(selective catalytic reduction，SCR)系统的尿素喷射，提出一种基于注意力机制（attention mechanism，AM）的卷积神经网络（convolutional neural networks，CNN)-长短时记忆网络(long shortterm memory，LSTM）模型预测柴油机NO_x排放的方法，根据柴油机NO_x生成机理和车辆实际道路测试采集的数据选取相关变量；使用AM-CNN模型提取特征，利用LSTM模型对提取的特征进行分析预测NO_x排放。结果表明：该混合模型对NO_x排放的预测精度较高，计算时间较少，平均绝对误差为5.307×10^-6，决定系数为0.932。根据预测模型中输入参数权重分析影响NO_x生成的关键因素，可以为优化柴油机燃烧过程提供参考。

关键词：NO_x排放；预测模型；AM-CNN-LSTM；深度学习；柴油机

Prediction of diesel engine NO_x emission based on AM-CNN-LSTM model

LIU Xing¹, ZHOU Shengkai¹, TIAN Linxia¹, DENG Xiaochao¹, LIN Penghui¹, LIU Zedu², LEI Yan^2*

1. Guangxi Yuchai Machinery Co., Ltd., Yulin 537000, China;

2. Vehicle Engineering Department, Department of Mechanical and Energy Engineering, Beijing University of Technology, Beijing 100124, China

Abstract：In order to accurately control the urea injection of the selective catalytic reduction (SCR) system, the research proposes a convolutional neural network(CNN)-long short term memory(LSTM) model based on the attention mechanism (AM), and applies it to predict diesel engine NO_xemissions. The relevant variables are selected based on the diesel engine NO_x generation mechanism and the data collects from actual vehicle road tests. The AM-CNN model is used to extract features, and the LSTM model is used to perform the extraction on the extracted features. The results show that the hybrid model has higher prediction accuracy for NO_xemissions, with less calculation time, an average absolute error of 5.307×10^-6, and a coefficient of determination of 0.932. Analyzing the key factors affecting NO_xgeneration based on the weight of the input parameters in the prediction model can provide a reference for optimizing the diesel engine combustion process.

Keywords: NO_xemission;prediction model; AM-CNN-LSTM; deep learning; diesel engine